Increasing parallelizing compiler efficiency using commutative functions

Abstract

The need for suitable software abstractions is proportional to the complexity of the underlying computer architecture. Even though language technology has made significant contributions over the years, many hard decisions have been delegated to the compiler. Naturally, compilers are only as efficient as the algorithms they contain. In this thesis we consider an approach to detecting commuting operations that give the same result independent of execution order. Our approached is based on analyzing commutativity of any pair of statements rather than just consecutive function calls. This will allow us to detect opportunities for parallelization that previous conservative approaches would ignore or miss be- cause of context-insensitivity.